My PhD project focuses on the monitoring of AI systems, specifically neural networks, to detect out of distribution inputs and unexpected behaviour. The goal is to create lightweight and accurate monitoring algorithms that work alongside machine learning models at deployment time in order to assure safety of the neural networks.
I chose to do my PhD with the STAI CDT because my research interests align with the aims of the CDT to promote safety within AI. During my years working in finance, I had firsthand experience with how a machine learning system fault could lead to substantial financial and reputational loss. There is also a tension between robustness and deployment speed; robustness was often achieved by the manual checking of system outputs which slows deployment speed. Thus, I was interested in exploring methods of assuring safety in machine learning without significantly affecting computational complexity.
The STAI CDT also allows me to be a member of a cohort of students with similar interests. This has been really valuable for idea generation and creating a social environment. Additionally, the training sessions and seminars have helped me develop a deeper understanding of the topics and techniques within safe and trusted AI.
—
Undergraduate Qualification: BA Computer Science, Harvard University
Masters Qualification: MSc Advanced Computer Science, Oxford University
Work Experience:
- Commodities Sales Strats Analyst, Goldman Sachs
- Foreign Exchange Options Quantitative Developer, NatWest Markets
Linkedin: https://www.linkedin.com/in/mariastoica/
